Re: gio, gvfs and nautilus fixes

[Alexander Larsson <alexl redhat com>]

On Mon, 2008-02-25 at 12:54 -0500, David Zeuthen wrote:

> > I cleaned up the completion code a bit to do less magic string uri
> > handling and instead use GFile apis.
> 
> It's a lot more simple now.. but you broke this important use-case
> 
>  $ gvfs-ls <tab>
>  file:///            sftp://hook.local/
>  $ gvfs-ls s<tab>
> 
> and it should complete to sftp://hook.local/

Well. what if there is a file starting with s? Anyway, fixing that
should be easy.

> Also, there's no completion on ./ or ~/. Typically ./ is used to
> complete on files in $CWD as completion on the empty string will give
> you the user visible mounts (which is a design choice). So some of the
> magic handling needs to be added back...

The actual gvfs-ls command handles expansion of all these when run
manually, including everything like ".." and combinations of these.
However, when started from bash completions it doesn't. I don't
understand why...

> > There was an unused "lock" introduced in GDaemonVolumeMonitor.
> > 
> > There was some races possible in the async find-mount code that i fixed.
> > 
> > Also, is there any particular reason the GDaemonVolumeMonitor lock is
> > recursive? I don't see why it would have to be?
> 
> I guess I was just playing it safe.

Recursive locks aren't imho "safe" for things like this. In most cases
recursive locks aren't needed or safe, and I then much prefer an easy to
read deadlock backtrace to some weird reentrancy happening that causes
rare crashers or other weird behaviour.